// Constants smaller than 256 fit in the immediate field of
// Thumb1 instructions so we return a zero cost and 1 otherwise.
int ARMTTIImpl::getIntImmCodeSizeCost(unsigned Opcode, unsigned Idx,
const APInt &Imm, Type *Ty) {
if (Imm.isNonNegative() && Imm.getLimitedValue() < 256)
return 0;
return 1;
}
/// We do not support symbolic projections yet, only 32-bit unsigned integers.
bool swift::getIntegerIndex(SILValue IndexVal, unsigned &IndexConst) {
if (auto *IndexLiteral = dyn_cast<IntegerLiteralInst>(IndexVal)) {
APInt ConstInt = IndexLiteral->getValue();
// IntegerLiterals are signed.
if (ConstInt.isIntN(32) && ConstInt.isNonNegative()) {
IndexConst = (unsigned)ConstInt.getSExtValue();
return true;
}
}
return false;
}
static bool isDereferenceableFromAttribute(const Value *BV, APInt Offset,
Type *Ty, const DataLayout &DL,
const Instruction *CtxI,
const DominatorTree *DT,
const TargetLibraryInfo *TLI) {
assert(Offset.isNonNegative() && "offset can't be negative");
assert(Ty->isSized() && "must be sized");
APInt DerefBytes(Offset.getBitWidth(), 0);
bool CheckForNonNull = false;
if (const Argument *A = dyn_cast<Argument>(BV)) {
DerefBytes = A->getDereferenceableBytes();
if (!DerefBytes.getBoolValue()) {
DerefBytes = A->getDereferenceableOrNullBytes();
CheckForNonNull = true;
}
} else if (auto CS = ImmutableCallSite(BV)) {
DerefBytes = CS.getDereferenceableBytes(0);
if (!DerefBytes.getBoolValue()) {
DerefBytes = CS.getDereferenceableOrNullBytes(0);
CheckForNonNull = true;
}
} else if (const LoadInst *LI = dyn_cast<LoadInst>(BV)) {
if (MDNode *MD = LI->getMetadata(LLVMContext::MD_dereferenceable)) {
ConstantInt *CI = mdconst::extract<ConstantInt>(MD->getOperand(0));
DerefBytes = CI->getLimitedValue();
}
if (!DerefBytes.getBoolValue()) {
if (MDNode *MD =
LI->getMetadata(LLVMContext::MD_dereferenceable_or_null)) {
ConstantInt *CI = mdconst::extract<ConstantInt>(MD->getOperand(0));
DerefBytes = CI->getLimitedValue();
}
CheckForNonNull = true;
}
}
if (DerefBytes.getBoolValue())
if (DerefBytes.uge(Offset + DL.getTypeStoreSize(Ty)))
if (!CheckForNonNull || isKnownNonNullAt(BV, CtxI, DT, TLI))
return true;
return false;
}
static Value *emit_untyped_intrinsic(jl_codectx_t &ctx, intrinsic f, Value **argvalues, size_t nargs,
jl_datatype_t **newtyp, jl_value_t *xtyp)
{
Value *x = nargs > 0 ? argvalues[0] : NULL;
Value *y = nargs > 1 ? argvalues[1] : NULL;
Value *z = nargs > 2 ? argvalues[2] : NULL;
Type *t = x->getType();
switch (f) {
case neg_int:
return ctx.builder.CreateNeg(x);
case add_int: return ctx.builder.CreateAdd(x, y);
case sub_int: return ctx.builder.CreateSub(x, y);
case mul_int: return ctx.builder.CreateMul(x, y);
case sdiv_int: return ctx.builder.CreateSDiv(x, y);
case udiv_int: return ctx.builder.CreateUDiv(x, y);
case srem_int: return ctx.builder.CreateSRem(x, y);
case urem_int: return ctx.builder.CreateURem(x, y);
// LLVM will not fold ptrtoint+arithmetic+inttoptr to GEP. The reason for this
// has to do with alias analysis. When adding two integers, either one of them
// could be the pointer base. With getelementptr, it is clear which of the
// operands is the pointer base. We also have this information at the julia
// level. Thus, to not lose information, we need to have a separate intrinsic
// for pointer arithmetic which lowers to getelementptr.
case add_ptr: {
return ctx.builder.CreatePtrToInt(
ctx.builder.CreateGEP(T_int8,
ctx.builder.CreateIntToPtr(x, T_pint8), y), t);
}
case sub_ptr: {
return ctx.builder.CreatePtrToInt(
ctx.builder.CreateGEP(T_int8,
ctx.builder.CreateIntToPtr(x, T_pint8), ctx.builder.CreateNeg(y)), t);
}
// Implements IEEE negate. See issue #7868
case neg_float: return math_builder(ctx)().CreateFSub(ConstantFP::get(t, -0.0), x);
case neg_float_fast: return math_builder(ctx, true)().CreateFNeg(x);
case add_float: return math_builder(ctx)().CreateFAdd(x, y);
case sub_float: return math_builder(ctx)().CreateFSub(x, y);
case mul_float: return math_builder(ctx)().CreateFMul(x, y);
case div_float: return math_builder(ctx)().CreateFDiv(x, y);
case rem_float: return math_builder(ctx)().CreateFRem(x, y);
case add_float_fast: return math_builder(ctx, true)().CreateFAdd(x, y);
case sub_float_fast: return math_builder(ctx, true)().CreateFSub(x, y);
case mul_float_fast: return math_builder(ctx, true)().CreateFMul(x, y);
case div_float_fast: return math_builder(ctx, true)().CreateFDiv(x, y);
case rem_float_fast: return math_builder(ctx, true)().CreateFRem(x, y);
case fma_float: {
assert(y->getType() == x->getType());
assert(z->getType() == y->getType());
Value *fmaintr = Intrinsic::getDeclaration(jl_Module, Intrinsic::fma, makeArrayRef(t));
return ctx.builder.CreateCall(fmaintr, {x, y, z});
}
case muladd_float: {
#if JL_LLVM_VERSION >= 50000
// LLVM 5.0 can create FMA in the backend for contractable fmul and fadd
// Emitting fmul and fadd here since they are easier for other LLVM passes to
// optimize.
auto mathb = math_builder(ctx, false, true);
return mathb().CreateFAdd(mathb().CreateFMul(x, y), z);
#else
assert(y->getType() == x->getType());
assert(z->getType() == y->getType());
Value *muladdintr = Intrinsic::getDeclaration(jl_Module, Intrinsic::fmuladd, makeArrayRef(t));
return ctx.builder.CreateCall(muladdintr, {x, y, z});
#endif
}
case checked_sadd_int:
case checked_uadd_int:
case checked_ssub_int:
case checked_usub_int:
case checked_smul_int:
case checked_umul_int: {
assert(x->getType() == y->getType());
Intrinsic::ID intr_id =
(f == checked_sadd_int ?
Intrinsic::sadd_with_overflow :
(f == checked_uadd_int ?
Intrinsic::uadd_with_overflow :
(f == checked_ssub_int ?
Intrinsic::ssub_with_overflow :
(f == checked_usub_int ?
Intrinsic::usub_with_overflow :
(f == checked_smul_int ?
Intrinsic::smul_with_overflow :
Intrinsic::umul_with_overflow)))));
Value *intr = Intrinsic::getDeclaration(jl_Module, intr_id, makeArrayRef(t));
Value *res = ctx.builder.CreateCall(intr, {x, y});
Value *val = ctx.builder.CreateExtractValue(res, ArrayRef<unsigned>(0));
Value *obit = ctx.builder.CreateExtractValue(res, ArrayRef<unsigned>(1));
Value *obyte = ctx.builder.CreateZExt(obit, T_int8);
jl_value_t *params[2];
params[0] = xtyp;
params[1] = (jl_value_t*)jl_bool_type;
jl_datatype_t *tuptyp = jl_apply_tuple_type_v(params, 2);
*newtyp = tuptyp;
Value *tupval;
tupval = UndefValue::get(julia_type_to_llvm((jl_value_t*)tuptyp));
tupval = ctx.builder.CreateInsertValue(tupval, val, ArrayRef<unsigned>(0));
tupval = ctx.builder.CreateInsertValue(tupval, obyte, ArrayRef<unsigned>(1));
return tupval;
}
case checked_sdiv_int: {
Value *typemin = ctx.builder.CreateShl(ConstantInt::get(t, 1), t->getPrimitiveSizeInBits() - 1);
raise_exception_unless(ctx,
ctx.builder.CreateAnd(
ctx.builder.CreateICmpNE(y, ConstantInt::get(t, 0)),
ctx.builder.CreateOr(
ctx.builder.CreateICmpNE(y, ConstantInt::get(t, -1, true)),
ctx.builder.CreateICmpNE(x, typemin))),
literal_pointer_val(ctx, jl_diverror_exception));
return ctx.builder.CreateSDiv(x, y);
}
case checked_udiv_int:
raise_exception_unless(ctx,
ctx.builder.CreateICmpNE(y, ConstantInt::get(t, 0)),
literal_pointer_val(ctx, jl_diverror_exception));
return ctx.builder.CreateUDiv(x, y);
case checked_srem_int:
return emit_checked_srem_int(ctx, x, y);
case checked_urem_int:
raise_exception_unless(ctx,
ctx.builder.CreateICmpNE(y, ConstantInt::get(t, 0)),
literal_pointer_val(ctx, jl_diverror_exception));
return ctx.builder.CreateURem(x, y);
case eq_int: *newtyp = jl_bool_type; return ctx.builder.CreateICmpEQ(x, y);
case ne_int: *newtyp = jl_bool_type; return ctx.builder.CreateICmpNE(x, y);
case slt_int: *newtyp = jl_bool_type; return ctx.builder.CreateICmpSLT(x, y);
case ult_int: *newtyp = jl_bool_type; return ctx.builder.CreateICmpULT(x, y);
case sle_int: *newtyp = jl_bool_type; return ctx.builder.CreateICmpSLE(x, y);
case ule_int: *newtyp = jl_bool_type; return ctx.builder.CreateICmpULE(x, y);
case eq_float: *newtyp = jl_bool_type; return math_builder(ctx)().CreateFCmpOEQ(x, y);
case ne_float: *newtyp = jl_bool_type; return math_builder(ctx)().CreateFCmpUNE(x, y);
case lt_float: *newtyp = jl_bool_type; return math_builder(ctx)().CreateFCmpOLT(x, y);
case le_float: *newtyp = jl_bool_type; return math_builder(ctx)().CreateFCmpOLE(x, y);
case eq_float_fast: *newtyp = jl_bool_type; return math_builder(ctx, true)().CreateFCmpOEQ(x, y);
case ne_float_fast: *newtyp = jl_bool_type; return math_builder(ctx, true)().CreateFCmpUNE(x, y);
case lt_float_fast: *newtyp = jl_bool_type; return math_builder(ctx, true)().CreateFCmpOLT(x, y);
case le_float_fast: *newtyp = jl_bool_type; return math_builder(ctx, true)().CreateFCmpOLE(x, y);
case fpiseq: {
*newtyp = jl_bool_type;
Type *it = INTT(t);
Value *xi = ctx.builder.CreateBitCast(x, it);
Value *yi = ctx.builder.CreateBitCast(y, it);
return ctx.builder.CreateOr(ctx.builder.CreateAnd(ctx.builder.CreateFCmpUNO(x, x),
ctx.builder.CreateFCmpUNO(y, y)),
ctx.builder.CreateICmpEQ(xi, yi));
}
case fpislt: {
*newtyp = jl_bool_type;
Type *it = INTT(t);
Value *xi = ctx.builder.CreateBitCast(x, it);
Value *yi = ctx.builder.CreateBitCast(y, it);
return ctx.builder.CreateOr(
ctx.builder.CreateAnd(
ctx.builder.CreateFCmpORD(x, x),
ctx.builder.CreateFCmpUNO(y, y)),
ctx.builder.CreateAnd(
ctx.builder.CreateFCmpORD(x, y),
ctx.builder.CreateOr(
ctx.builder.CreateAnd(
ctx.builder.CreateICmpSGE(xi, ConstantInt::get(it, 0)),
ctx.builder.CreateICmpSLT(xi, yi)),
ctx.builder.CreateAnd(
ctx.builder.CreateICmpSLT(xi, ConstantInt::get(it, 0)),
ctx.builder.CreateICmpUGT(xi, yi)))));
}
case and_int: return ctx.builder.CreateAnd(x, y);
case or_int: return ctx.builder.CreateOr(x, y);
case xor_int: return ctx.builder.CreateXor(x, y);
case shl_int:
return ctx.builder.CreateSelect(
ctx.builder.CreateICmpUGE(y, ConstantInt::get(y->getType(),
t->getPrimitiveSizeInBits())),
ConstantInt::get(t, 0),
ctx.builder.CreateShl(x, uint_cnvt(ctx, t, y)));
case lshr_int:
return ctx.builder.CreateSelect(
ctx.builder.CreateICmpUGE(y, ConstantInt::get(y->getType(),
t->getPrimitiveSizeInBits())),
ConstantInt::get(t, 0),
ctx.builder.CreateLShr(x, uint_cnvt(ctx, t, y)));
case ashr_int:
return ctx.builder.CreateSelect(
ctx.builder.CreateICmpUGE(y, ConstantInt::get(y->getType(),
t->getPrimitiveSizeInBits())),
ctx.builder.CreateAShr(x, ConstantInt::get(t, t->getPrimitiveSizeInBits() - 1)),
ctx.builder.CreateAShr(x, uint_cnvt(ctx, t, y)));
case bswap_int: {
Value *bswapintr = Intrinsic::getDeclaration(jl_Module, Intrinsic::bswap, makeArrayRef(t));
return ctx.builder.CreateCall(bswapintr, x);
}
case ctpop_int: {
Value *ctpopintr = Intrinsic::getDeclaration(jl_Module, Intrinsic::ctpop, makeArrayRef(t));
return ctx.builder.CreateCall(ctpopintr, x);
}
case ctlz_int: {
Value *ctlz = Intrinsic::getDeclaration(jl_Module, Intrinsic::ctlz, makeArrayRef(t));
y = ConstantInt::get(T_int1, 0);
return ctx.builder.CreateCall(ctlz, {x, y});
}
case cttz_int: {
Value *cttz = Intrinsic::getDeclaration(jl_Module, Intrinsic::cttz, makeArrayRef(t));
y = ConstantInt::get(T_int1, 0);
return ctx.builder.CreateCall(cttz, {x, y});
}
case abs_float: {
Value *absintr = Intrinsic::getDeclaration(jl_Module, Intrinsic::fabs, makeArrayRef(t));
return ctx.builder.CreateCall(absintr, x);
}
case copysign_float: {
Value *bits = ctx.builder.CreateBitCast(x, t);
Value *sbits = ctx.builder.CreateBitCast(y, t);
unsigned nb = cast<IntegerType>(t)->getBitWidth();
APInt notsignbit = APInt::getSignedMaxValue(nb);
APInt signbit0(nb, 0); signbit0.setBit(nb - 1);
return ctx.builder.CreateOr(
ctx.builder.CreateAnd(bits, ConstantInt::get(t, notsignbit)),
ctx.builder.CreateAnd(sbits, ConstantInt::get(t, signbit0)));
}
case flipsign_int: {
ConstantInt *cx = dyn_cast<ConstantInt>(x);
ConstantInt *cy = dyn_cast<ConstantInt>(y);
if (cx && cy) {
APInt ix = cx->getValue();
APInt iy = cy->getValue();
return ConstantInt::get(t, iy.isNonNegative() ? ix : -ix);
}
if (cy) {
APInt iy = cy->getValue();
return iy.isNonNegative() ? x : ctx.builder.CreateSub(ConstantInt::get(t, 0), x);
}
Value *tmp = ctx.builder.CreateAShr(y, ConstantInt::get(t, cast<IntegerType>(t)->getBitWidth() - 1));
return ctx.builder.CreateXor(ctx.builder.CreateAdd(x, tmp), tmp);
}
case ceil_llvm: {
Value *ceilintr = Intrinsic::getDeclaration(jl_Module, Intrinsic::ceil, makeArrayRef(t));
return ctx.builder.CreateCall(ceilintr, x);
}
case floor_llvm: {
Value *floorintr = Intrinsic::getDeclaration(jl_Module, Intrinsic::floor, makeArrayRef(t));
return ctx.builder.CreateCall(floorintr, x);
}
case trunc_llvm: {
Value *truncintr = Intrinsic::getDeclaration(jl_Module, Intrinsic::trunc, makeArrayRef(t));
return ctx.builder.CreateCall(truncintr, x);
}
case rint_llvm: {
Value *rintintr = Intrinsic::getDeclaration(jl_Module, Intrinsic::rint, makeArrayRef(t));
return ctx.builder.CreateCall(rintintr, x);
}
case sqrt_llvm: {
Value *sqrtintr = Intrinsic::getDeclaration(jl_Module, Intrinsic::sqrt, makeArrayRef(t));
return ctx.builder.CreateCall(sqrtintr, x);
}
default:
assert(0 && "invalid intrinsic");
abort();
}
assert(0 && "unreachable");
}
